The ALS Association is committed to helping improve clinical trial design, in order to increase trial efficiency that will more quickly lead to effective therapeutics.
We awarded Dr. David Ennist and colleagues at Origent Data Sciences, Inc. two grants to support research exploring how machine learning algorithms, a type of computational tool, can optimize clinical trial design. Dr. Ennist’s work, recently published in the journal Annals of Clinical Trial and Translational Neurology, looks closely at optimizing patient randomization into clinical trials.
In addition to Origent colleagues, the paper was published with Dr. James Berry (Massachusetts General Hospital), Lisa Meng, Pharm.D., and Amy Bian (Cytokinetics), Dr. Jinsy Andrews (Columbia University), and Dr. Bernard Ravina (Voyager Therapeutics).
“ALS is a complex disease that presents and progresses differently for each person,” stated Dr. Lucie Bruijn, chief scientist at The ALS Association. “As a result, randomizing people into ALS clinical trials has proven difficult, especially in small trials where differences in patient characteristics across treatment arms can cause a trial to fail. The Association is delighted to support the innovative work of David Ennist and his team to address these challenges.”
“We’re grateful for the funding from The ALS Association. This funding has allowed us to validate our algorithms using data from recent, contemporary clinical trials in ALS – a crucial step in ensuring the accuracy and relevance of the models to current clinical trials,” stated Dr. Ennist.
Having accurate predictors of both function and survival for clinical trials would allow groups of patients to be stratified by a specific characteristic, so a new treatment could be tested separately in groups with that specific characteristic. This process could reduce the “noise” of clinical measurements and allow a treatment benefit “signal” to emerge more clearly.
Origent Data Sciences, Inc. has developed computational tools, called algorithms, to make those predictions using machine learning. Here, they applied these algorithms to improve clinical trial randomization. They decided to use a score of predicted survival as a single stratifier, or classifier.
To test their question whether their algorithm could improve clinical trial stratification, Origent used data from the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) Database to virtually simulate trials.
The PRO-ACT database is powerful because it contains more than 10 million data points extracted from the de-identified records of 10,700 patients who participated in 23 ALS Phase II & III clinical trials. Using PRO-ACT data avoided the need to use ongoing real-life clinical trial data.
Ennist and team randomized the virtual trials in two ways: 1) using their algorithm stratification method focusing on predicted survival; 2) using a more traditional stratification method focusing on riluzole use and bulbar onset. Then they analyzed treatment outcomes between the two groups. They also partnered with Cytokinetics under another grant from The ALS Association to gain access to the dataset from the placebo arm of BENEFIT-ALS, the completed phase IIb trial of tirasemtiv, to validate their findings. In addition to modeling the use of the predictions for randomizing the clinical trials, the team also used the predictions to control for confounders of the outcome in the backend analysis of the virtual trials.
They found that stratification by their predicted survival algorithm demonstrated fewer failures compared to traditional stratification. Importantly, the backend analysis showed that a substantial decrease in sample size (number of patients in a trial) was needed to achieve statistical power using their predicted survival algorithm. These methods can also be used to increase the power of a clinical study, thus making it more likely that a given study will be successful.
The potential impact on ALS clinical trial design is immense. This new stratification method could result in faster, less expensive ALS clinical trials that enroll fewer patients, have a greater chance of succeeding, and result in a quicker path to new ALS therapeutics.
We applaud Dr. Ennist and his team and look forward to learning how they will continue to apply their algorithmic models to optimize clinical trial designs for not only ALS, but also similar neurological diseases.
“We discussed this approach last December with the FDA and they have advised us to apply it to increasing the power of a clinical trial. We’re looking forward to working with drug developers to apply this technology in upcoming ALS clinical trials,” stated Dr. Ennist.
In addition to the funding from The ALS Association, the work was partially funded by Voyager.
Improved stratification of ALS clinical trials using predicted survival. Berry J., et al. Annals of Clinical and Translational Neurology. Published online: March 9, 2018.
For open access article: https://doi.org/10.1002/acn3.550